This paper proposes novel spatial feature extraction combining wavelet analysis and sparse matrix techniques to solve the problem of identifying objects that are of subtle differences. This hybrid matrix feature is not put forward before in any literature. The differences between slightly dissimilar objects are distinctions in the spatial orientations of the objects or the local positions of points on their contours. The time-frequency localization of wavelet transform distinguishes these differences and leads to a sparse form of underlying objects. This sparsity allow us re-arrange the pixels in the wavelet decomposed details sub-images. Treating three directional details as sparse matrices, different sparse matrix reordering are applied upon them. The reordering produces a considerable increase of the distinction between slightly dissimilar objects. In consequence, the difficulty to discriminate between objects is largely reduced. A series of discriminative simulations are shown which verify the feasibility and effectiveness of the proposed method.